Hi,
I tried to set up a cluster as you suggested.
First, I started four instances with the commands below.
ec2-add-group "_at_sc-cluster1" -d "StarCluster cluster1"
ec2-authorize "_at_sc-cluster1" -P tcp -p 22 -s 0.0.0.0/0
ec2-request-spot-instances ami-a19e71c8 --group "_at_sc-masters" --group
"_at_sc-cluster1" -m -z us-east-1a -p 1.0 -t c1.xlarge -n 1 -r one-time
ec2-request-spot-instances ami-a19e71c8 --group "_at_sc-cluster1" -m -z
us-east-1a -p 1.0 -t c1.xlarge -n 3 -r one-time
starcluster start -x high-cpu-xl cluster1
Once the instances were running, I ran the command "starcluster start
-x high-cpu-xl cluster1". However, it did not go beyond the text
"Waiting for cluster to start...".
What could I be doing wrong?
- Joonas
Here's my config for the cluster:
[cluster small]
KEYNAME = pstam-keypair
CLUSTER_SIZE = 1
CLUSTER_USER = jpaalasm
CLUSTER_SHELL = bash
NODE_IMAGE_ID = ami-8f9e71e6
NODE_INSTANCE_TYPE = m1.small
AVAILABILITY_ZONE = us-east-1a
VOLUMES = shhs
[cluster high-cpu-xl]
EXTENDS = small
NODE_IMAGE_ID = ami-a19e71c8
NODE_INSTANCE_TYPE = c1.xlarge
On Wed, Mar 24, 2010 at 5:20 PM, Justin Riley <jtriley_at_mit.edu> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Hi Joonas,
>
> The development code in github for the next version can use spot
> instances provided that you launch them manually.
>
> The next version of starcluster makes this possible because of the fact
> that it relies on creating/using security groups for accounting.
>
> By default StarCluster creates a security group for all starcluster
> master nodes called _at_sc-masters. When you launch a cluster, say called
> 'mycluster', StarCluster creates a new security group called
> _at_sc-mycluster. It then runs the master and assigns it to both
> _at_sc-masters and @sc-mycluster security groups and then runs the rest of
> the nodes, assigning them only to _at_sc-mycluster.
>
> So, with the new version, you can manually create a _at_sc-mycluster
> security group, launch your instances by adding the master to both the
> _at_sc-masters AND @sc-mycluster security groups and then launching the
> rest of the nodes in the _at_sc-mycluster security group only. Starcluster
> will then be able to launch using these instances, provided they're in a
> pending/running state, by running the following:
>
> $ starcluster start -x smallcluster mycluster
>
> This will then skip requesting new instances and wait for all existing
> instances in the _at_sc-mycluster group to come up and it will continue on
> normally from there.
>
> Does this make sense? I will document this in the next version. If
> you're interested in using the development version you will find docs
> for the new config in the github code (starcluster/docs/sphinx/*.rst).
>
> Hope that helps,
>
> ~Justin
>
>
>
>
> On 03/24/2010 04:33 AM, Joonas Paalasmaa wrote:
>> Hello,
>>
>> Are there any plans to support EC2 spot instances in starcluster?
>> Could it, for example, be implemented so that you specify the instance
>> IDs of of the launched StarCluster spot instances and starcluster
>> takes care of configuring a cluster from them?
>>
>> Thanks,
>> Joonas Paalasmaa
>> _______________________________________________
>> Starcluster mailing list
>> Starcluster_at_mit.edu
>> http://mailman.mit.edu/mailman/listinfo/starcluster
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v2.0.14 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
>
> iEYEARECAAYFAkuqLbkACgkQ4llAkMfDcrn4nQCeI1y57Rj8k9chXgFMe2KNhfXh
> WqIAnA7zP9Mb4nuRbrp5CP54mxtxb/mY
> =C1+x
> -----END PGP SIGNATURE-----
>
Received on Tue Mar 30 2010 - 09:00:24 EDT